Skip to content

Appendix C References

Citations in this book come in two kinds: in-vault notes (entity/concept/comparison/query pages of the Hermes Wiki knowledge base at ~/wiki, which themselves cite original external sources) and public sources (papers, official blogs, public reports). Convention: for data marked "via in-vault notes," the original source cited by the note is the final authority. In-vault paths are searchable in Hermes Wiki.

C1 Core in-vault notes, by theme

Harness engineering

  • entities/harness-engineering — the paradigm's main entity: Agent = Model + Harness; three generations; six-layer; seven anti-patterns; 5 artifacts/3 camps/5 consensus principles; decay and Build to Delete; the sovereignty line; Can.ac and $9/$200 data; eight source threads including Alibaba Cloud's multi-agent practice and Ruo Fei's new context rules
  • concepts/harness-engineering-framework — the framework page: six-layer structure, five context tiers, Generator-Evaluator, compaction vs reset
  • topics/agent-harness-deep-dive-qa — the L1–L5 architecture, tool-layer mapping matrix, loop six primitives, the production system, open problems
  • queries/harness-minimum-checklist — MVH sides A/B, tool token accounts, bottleneck trio, shelf-life table, migration checklist
  • concepts/when-not-to-harness-engineering — the opposing four boundary conditions, four questions, the NomShub escape chain
  • comparisons/model-capability-vs-harness-engineering — the adversarial page, layered temporality verdict, three testable predictions
  • Context-vs-Harness diagnostic table (in entities/harness-engineering, source 6)

Loop engineering

  • moc/loop-engineering — the topic map: ten core theses, the five-stage cycle, Loop≠Cron, cost structure
  • entities/loop-engineering-feedback-control-system — the main entity: closed-loop four, six components, single vs fleet, Samuel McDonnell's critique, the Bun case, control-theory mapping
  • comparisons/prompt-engineering-vs-context-engineering, comparisons/vibe-coding-vs-agentic-engineering — paradigm contrasts

Graph engineering

  • entities/langgraph-state-machine-under-the-hood — the triad, reducers, compile, fan-out/fan-in, stream
  • comparisons/orchestrator-worker-vs-dag-agent, comparisons/single-agent-vs-multi-agent — topologies and migration criteria
  • topics/multi-agent-systems — collaboration pattern spectrum, JSONL inboxes, summary return, Pass@k/Pass^k

Memory and context

  • topics/agent-memory-systems — memory types, four modeled objects, three-way retrieval, lifecycle governance
  • queries/why-agent-poc-fails-production — six failure categories, ten-item list, the 81% review-overhead data, Tencent's knowledge tiers

Paradigm and history

  • drafts/karpathy-2026-vibe-to-agentic-engineering — a cross-cut of 184 notes: vibe→agentic evolution, L1–L6, animals vs ghosts, industry echo
  • entities/vibe-coding-god-object-7months-failure, entities/tencent-vibe-coding-to-agentic-engineering-backend — the crash post-mortem and the team's retrospective

C2 Public sources

Papers

  • Vaswani et al., Attention Is All You Need. NeurIPS 2017. arXiv:1706.03762
  • Yao et al., ReAct: Synergizing Reasoning and Acting in Language Models. ICLR 2023. arXiv:2210.03629
  • Wei et al., Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. NeurIPS 2022. arXiv:2201.11903
  • Shinn et al., Reflexion: Language Agents with Verbal Reinforcement Learning. 2023. arXiv:2303.11366
  • Liu et al., Lost in the Middle: How Language Models Use Long Contexts. TACL 2024. arXiv:2307.03172
  • Stanford / UC Berkeley / NVIDIA, LLM-as-a-Verifier. 2026 (Terminal-Bench 2.0 data; via drafts/karpathy-2026-vibe-to-agentic-engineering)
  • METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. 2025

Engineering blogs and official documents

  • Anthropic, Effective Harnesses for Long-Running Agents. 2025-11
  • Anthropic, When AI builds itself. 2026-06
  • Anthropic, MCP (Model Context Protocol) announcement and documentation. From 2024-11
  • OpenAI (Ryan Lopopolo), Harness engineering: leveraging Codex in an agent-first world. 2026-02
  • Mitchell Hashimoto, My AI Adoption Journey. 2026-02
  • Ben Thompson, Agents Over Bubbles. Stratechery. 2026-03
  • Stripe Engineering (Alistair Gray), Minions: Stripe's one-shot, end-to-end coding agents. 2026-02
  • LangChain (Vivek Trivedy), Improving Deep Agents with harness engineering. 2026-02
  • Birgitta Böckeler (martinfowler.com), Harness Engineering. 2026-02
  • Can.ac, I Improved 15 LLMs at Coding in One Afternoon. Only the Harness Changed. 2026-02
  • GitHub Blog, How to write a great AGENTS.md: lessons from over 2,500 repositories
  • Addy Osmani, Loop Engineering (the six-primitive framework). 2026
  • Google, A2A (Agent2Agent Protocol). 2025-04 (later donated to the Linux Foundation)
  • LangGraph official documentation (StateGraph / Reducer / Checkpointer)
  • LinearB annual engineering-efficiency report (AI PR rejection rates). 2025

C3 Citation and currency statement

  1. Every quantitative claim in this book is sourced at its point of use. Different sources may report different numbers for the same phenomenon (measurement differences); where they do, this book prefers the value explicitly recorded in the cited in-vault note and says so.
  2. Harness/Loop/Graph evolve quickly: tool capabilities and data are current as of mid-2026. The book is revised continuously alongside the knowledge base's daily check & eval process.
  3. On contested questions (model capability vs harness engineering), both sides' evidence and the verdict framework are presented (Ch 2, Ch 13); the book does not conclude by allegiance.
  4. Found an error? Open an issue in this repository; corrections ship in the next edition and are annotated in the affected chapters.